%cst = arith.constant dense<1.0> : tensor<4xf32>
  %cst_3 = arith.constant dense<2.0> : tensor<4x3x3x3xf32>
  %0 = "tfl.conv_2d"(%arg0, %cst_3, %cst) {dilation_h_factor = 1 : i32, dilation_w_factor = 1 : i32, fused_activation_function = "RELU6", padding = "VALID", stride_h = 2 : i32, stride_w = 2 : i32} : (tensor<?x19x19x3xf32>, tensor<4x3x3x3xf32>, tensor<4xf32>) -> tensor<?x9x9x4xf32>
  func.return %0 : tensor<?x9x9x4xf32>

};

TEST_F(VerifyNoOutsideCompilationMarkersPassTest, PassesValidOps) {
  static constexpr char kMlirModuleStr[] = R"(
  module attributes {tf.versions = {bad_consumers = [], min_consumer = 0 : i32, producer = 268 : i32}} {
    func.func @main() -> tensor<1xi32> {
      %0 = "tf.Const"() {value = dense<1000> : tensor<1xi32>} : () -> tensor<1xi32>
      func.return %0 : tensor<1xi32>
    }
  })";

  // CHECK-NEXT: }, {
  }, {
    // CHECK-NEXT: tf_device.return
    tf_device.return
  // CHECK-NEXT: }) : () -> (tensor<i1>, tensor<i64>, tensor<f64>)
  }) : () -> (tensor<i1>, tensor<i32>, tensor<i64>, tensor<f32>, tensor<f64>)
  "tf.AssignVariableOp"(%arg5, %0#1) : (tensor<!tf_type.resource>, tensor<i32>) -> ()

      Targuments = [!tf_type.resource, !tf_type.resource],
      Tstate = [i64, i32], device = "",
      f = @__reduce_func_3, f._tf_data_function = true,
      output_shapes = [#tf_type.shape<>, #tf_type.shape<>],

  func.func @main() -> () {
    func.return
  }

  module @submodule attributes {tf.devices = {"/job:localhost/replica:0/task:0/device:CPU:0", "/job:localhost/replica:0/task:0/device:TPU:0", "/job:localhost/replica:0/task:0/device:TPU:1", "/job:localhost/replica:0/task:0/device:TPU_SYSTEM:0"}, tf.versions = {bad_consumers = [], min_consumer = 12 : i32, producer = 1647 : i32}} {
  }

// CHECK: %[[CST:.*]] = "tfl.no_value"() <{value}> : () -> none
// CHECK: %[[RES0:.*]] = "tfl.conv_2d"(%arg0, %arg1, %[[CST]]) <{dilation_h_factor = 1 : i32, dilation_w_factor = 1 : i32, fused_activation_function = "NONE", padding = "SAME", stride_h = 1 : i32, stride_w = 1 : i32}> : (tensor<256x32x32x3xf32>, tensor<16x3x3x3xf32>, none) -> tensor<256x32x32x16xf32>
// CHECK: return %[[RES0]] : tensor<256x32x32x16xf32>

{
  "version": 3,

// CHECK-LABEL @export_three
module @export_three attributes {tf_saved_model.semantics} {
  // CHECK: func @a
  // CHECK-SAME: i32
  // CHECK-SAME: f32
  // CHECK-SAME: tensor<i32>
  // CHECK: [[tuple:%.*]]:2 = call @f
  // CHECK: return [[tuple]]#0, [[tuple]]#1
  // CHECK-SAME: tensor<i8>
  // CHECK-SAME: tensor<i16>
  // CHECK: func @b
  // CHECK: func @c

  // Using a string constant here instead of testdata to make this compatible
  // with open source.
  static constexpr char kMlirModuleStr[] = R"(
  module attributes {tf.versions = {bad_consumers = [], min_consumer = 0 : i32, producer = 268 : i32}} {
    func.func @main() -> tensor<1xi32> {
      %0 = "tf.BadValue"() {value = dense<1000> : tensor<1xi32>} : () -> tensor<1xi32>
      func.return %0 : tensor<1xi32>
    }
  })";

      (tensor<i1>, tensor<i1>, tensor<3x15x14x3xf32>, tensor<3x15x14x3xf32>) -> tensor<i1>
  %4 = "tfl.conv_2d"(%arg0, %cst_2, %cst_1) {dilation_h_factor = 1 : i32, dilation_w_factor = 1 : i32, fused_activation_function = "NONE", padding = "SAME", stride_h = 1 : i32, stride_w = 1 : i32} : (tensor<3x15x14x3xf32>, tensor<8x3x3x3xf32>, tensor<8xf32>) -> tensor<3x15x14x8xf32>
  func.return %4 : tensor<3x15x14x8xf32>
}

  // CHECK-NEXT: mhlo.constant
  %0 = mhlo.constant dense<-1> : tensor<i32>
  %1 = mhlo.constant dense<0> : tensor<i32>
  %2 = mhlo.constant dense<20> : tensor<i32>
  // CHECK-NEXT: mhlo.while
  %3, %4, %5 = mhlo.while(%iterArg = %1, %iterArg_0 = %0, %iterArg_1 = %1) : tensor<i32>, tensor<i32>, tensor<i32>
   cond {
    %17 = mhlo.compare  LT, %iterArg_1, %2 : (tensor<i32>, tensor<i32>) -> tensor<i1>
    mhlo.return %17 : tensor<i1>
  } do {